A comparative study of machine learning based automatic diagnosis approaches to diagnose knee osteoarthritis from radiographs.

Background

Knee Osteoarthritis (KOA) is a common degenerative joint condition that affects millions of middle-aged and elderly people worldwide, mainly due to the gradual loss of articular cartilage. While diagnosis often depends on X-ray imaging, grading KOA severity with Kellgren-Lawrence (KL) scales is subjective and varies between observers.

Purpose

To address these issues, machine learning (ML) based computer-aided diagnosis (CAD) systems have become valuable tools for automating KOA detection and grading. This review offers a detailed comparative analysis of traditional and deep learning ML methods developed over the last decade, focusing on models trained with radiographic (X-ray) images the most accessible and cost-effective modality for KOA evaluation.

Methods

We trace the evolution from handcrafted feature-based models to end-to-end deep learning architectures such as Convolutional Neural Networks (CNNs), Inception models, Residual Networks (ResNets), and Transformers, emphasizing key datasets, methodological advances, and diagnostic accuracy.

Results

AND

Conclusions

The review discusses challenges like data imbalance, model generalization, and clinical interpretability, along with future research directions. Overall, this review aims to help researchers and clinicians understand the current state of KOA diagnosis and develop robust, scalable, and clinically applicable systems.